GMDH based back propagation algorithm to predict abutment scour in cohesive soils

Abstract

In this study, group method of data handling (GMDH) network is used to predict abutments scour depth of bridges which have been embedded in two types of montmorillonite and kaolinite clay soils. The GMDH network has been developed using back propagation technique. Several effective parameters including initial water content, clay content, degree of compaction, and non-dimensional shear strength of bed soils were derived using dimensional analysis for modeling of abutment scour depth. Training and testing results of the GMDH network were compared with those performed using adaptive neuro-fuzzy inference system (ANFIS), radial basis function-neural network (RBF-NN), and traditional equations. Also, efficiency of the GMDH–BP was investigated by different classifications of initial water content (IWC) and degree of compaction (CC) ranges. Results showed that the GMDH–BP had the higher performance for unsaturated montmorillonite clay with IWC≤25%. From the results, it was perceived that this method emerged as the most accurate as ANFIS, RBF-NN, and traditional equations. In particular application, the GMDH network was presented as a new technique for prediction of scour depth around abutment in cohesive bed materials.